AI-Based Anomaly Detection in Cloud Computing Environments Using Ensemble Learning

Abstract

Cloud computing environments generate continuously changing network and performance data, making anomaly detection essential for identifying intrusions, service degradation, and abnormal resource behavior. This paper examines AI-based anomaly detection in cloud computing with emphasis on ensemble learning. A structured comparative evidence analysis integrates verified peer-reviewed results from intrusion-detection and cloud-monitoring studies using KDDCup99, NSL-KDD, UNSW-NB15, CICIDS2017, DApp monitoring data, Server Machine Dataset, and Vichalana data. The comparison focuses on accuracy, precision, recall, F1 score, class-imbalance handling, robustness, and short-horizon prediction. Published evidence shows that ensemble methods often outperform individual classifiers under the same experimental conditions. On UNSW-NB15, an ensemble machine-learning model achieved 97.06% binary-detection accuracy with 98.45% F1, exceeding the individual Decision Tree, XGBoost, Random Forest, and Gradient Boosting models evaluated in the same study. The evidence indicates that ensembles strengthen anomaly detection through classifier diversity, variance reduction, error-focused learning, and imbalance handling. However, benchmark scores remain dependent on dataset realism, preprocessing, class distribution, and evaluation protocol. Reliable cloud deployment therefore requires representative data, multi-metric evaluation, and continuous validation under changing traffic and workload conditions